Channel Prediction For Massive MIMO Based On Deep Learning

In recent years,with the rapid development of mobile communication,5G wireless technology playing an importa role in communication domain.Massive multiple-input multiple-output(MIMO)is one of the key technologies of 5G and future communication because of great advantages in energy efficiency,equipment cost and anti-interference ability,and it can creates more capacity and transmission rate.massive MIMO will get greater development and wider application.Channel research is an important part of massive MIMO systems.When using Time Division Duplexing(TDD)mode,due to the reciprocity of upstream and downstream channels,We need to confirm the current channel state information(CSI)at the base station for the precoding and beam shaping of the downlink Channel.However,in the process of mobile communication,there will be some difference between the uplink Channel and downlink Channel with the movement of users or objects in the Channel.Which causes the channel aging problem.The traditional CSI acquisition method at the base station is mostly based on channel estimation,which is mostly based on orthogonal pilot.The MIMO channel data is not completely regarded as “big data” from the perspective of data analysis.And the channel estimation method is difficult to obtain CSI in real time and accurately under the condition of high dimension and complexity of CSI.Therefore,we considered using channel prediction method to solve this problem.Deep learning theory has injected new vitality into channel prediction research,and it has great potential to extend the deep learning method to channel prediction of massive MIMO systems.Therefore,this paper studies the channel prediction of massive MIMO based on deep learning.First,In this paper,the system model of massive MIMO communication system is introduced firstly,such as analyze the channel characteristics of massive MIMO system.In view of the channel time-varying characteristics of different scenarios and different moving speeds,the channel aging problem that needs to be solved in massive MIMO communication system is introduced.By investigating the advantages of data-driven methods based on “big data”,and this paper explains why deep learning is used to solve the channel prediction problem.The convolutional neural network(CNN)and recurrent Neural Network(RNN)and the variants which named long shot time memory(LSTM)are introduced emphatically.At the same time,COST2100 wireless channel model and measured data are used to generate the data set used in this paper.Second,we studied the channel prediction method of massive MIMO system in different scenarios.By analyzing the spatial and temporal correlation of channel state information,we proposed the construction idea of prediction model and a channel prediction model based on CNN and LSTM.The channel state information of MIMO systems with spatial,frequency and time domain features is processed by combining the characteristics of CNN and LSTM network which learning with long correlation sequence prediction.In the deeplearing algorithm model,3D convolutional neural network(3D CNN)was used to learn the local features of the channel state information,and a Convolution Long ShortTerm Memory(Conv LSTM)was used as a predictor.Finally,the feature recovery was completed by 2D CNN.At the same time,the simulation data set based on COST2100 wireless communication model is used to pre-train the neural network.On the basis of the pre-training model,the measured channel data set is used for auxiliary training of the model,so that the model has better generalization ability on the measured data set.Finally,we used Normalized Mean Square Error(NMSE)and sum-rate of multi-user MIMO downlink channel based on linear precoding to evaluate the prediction performance of channel prediction model based on deep learning.It is compared with autoregressive model(AR).The prediction accuracy is improved at least 15% than AR,and the system performance is improved at least 10% than AR under multiple linear precoding schemes.Therefore,the prediction method based on deep learning proposed in this paper is superior to AR prediction method on the whole.